Beruflich Dokumente
Kultur Dokumente
R3
1K
CG
100PF 2
STARTUP
CIRCUITRY
VIN
PWL
V(3)
RGC START
1K
VCC RBU
8V 2.2K
CCO
12PF
LBX
3NH
14
11
CBC
22PF
LS2
3NH
22
CB1 10PF
7
9 LB1
3NH
V(7)
CB2
VPOWER 47PF
LB2
LCX
3NH
5NH 6
XLSP
B9112
X1
QN5109
13
16
5
RBL
1K
15
V(15)
RL VOUT
50
R2 10E6
X2
DN5441
LS1
3NH
LT
20NH
18
RTI 10K
LEC 170NH
LEX
5NH
12
D1
DN4148
17
VVT
4V
V(3)
START
4
3
CEX
5PF
30
XLEC
B9112
REX
47
VARACTOR
Figure 1, A 500 MHz oscillator and RF Bead (shaded area) test circuit.
Page 1
Modeling An RF Bead
180
180
140
140
100.0
60.0
20.0
Z (2743009112) in Ohms
Z (2743003112) in Ohms
2
100.0
3
4
60.0
20.0
1MEG
10MEG
100MEG
1G
Figure 2, The Intusoft RF Bead model simulates the proper impedance vs.
frequency characteristics. The graph displays the response for several devices.
Starting An Oscillator
Simulation of oscillators present a variety of challenges, not the
least of which is getting the oscillator to oscillate. When ISSPICE
performs an AC or Transient analysis it first performs a DC
analysis in order to establish the starting initial operating point for
the circuit. If a stable operating point is found, which is the goal of
the DC analysis, the oscillator may not oscillate during the
transient (time domain) analysis unless some random disturbance is encountered.
There are a number of ways to start an oscillator; each with
varying results and consequences (Figure 3). The method chosen here was to introduce a voltage pulse into the circuit, specifically, at the emitter of the transistor (VIN 1 0 PWL...). Another
possible method is to insert a current pulse somewhere in the
resonant portion of the circuit, for example at LT (I1 18 17 PULSE
.01 0). If the DC analysis does not converge, a sign that the circuit
Page 2
5
5.0000N
15.000N
25.000N
35.000N
45.000N
TIME in Secs
is unstable and may want to oscillate without any help, the UIC
keyword can be issued in the .TRAN statement. For example,
.TRAN .1NS 50NS UIC. This will cause the simulation to proceed
directly to the Transient analysis bypassing the DC analysis. The
.IC and IC= parameters can then be used to set initial transient
conditions and unbalance the oscillator. One problem with using
UIC is that no DC operating point will be produced inhibiting study
of the circuit bias. The last method is similar to the first and
involves ramping of the power supply (VCC 8 0 PULSE 0 8V 0
5NS). This method may not work well, however, due to the bypass
capacitors. In general, when ramping a source, make sure to give
the ramp a realistic slope in order to avoid timestep too small
errors. The first two methods are the most often recommended as
the other methods may not work or may introduce transient startup residues [3].
Circuit Modeling
The oscillator circuit was first simulated with two inductor chokes,
LSP and LEC (Figure 1), and a standard .MODEL statement for
the 2N5109 transistor. In SPICE, the standard representation for
a transistor uses the Gummel-Poon model. Various parameters
in the SPICE .MODEL statement are altered in order to cause the
generic nature of the Gummel-Poon template to represent a
particular device.
In a second simulation, the chokes were each replaced with the
new RF bead model. The simple transistor model was replaced
with a subcircuit representation. Since the Gummel-Poon model
can not adequately represent BJT behavior above approximately
200MegHz, a composite model must be assembled. The SPICE
subcircuit, containing a BJT model and various parasitic elements, is utilized for this purpose.
Page 3
When simulating in SPICE it is best to use a subcircuit representation for a device rather than forcing model parameters to have
unreasonable values. If model parameters are used outside their
physical bounds, the model may work well in one area, but
incorrectly in another. For example, the device may behave
properly during the AC small signal analysis, but poorly during the
nonlinear transient analysis. Some vendors who produce SPICE
model use this approach and the user should beware. Models for
RF transistors that are going to be used for both linear and
nonlinear analysis can not be produced this way. A subcircuit
representation must be used!
Note: Parasitics for the passive elements were maintained for
both simulation cases.
1.000
14.69
-1.000
-2.000
12.69
VPOWER in Volts
VOUT in Volts
10.69
8.690
2
-3.000
6.690
5.000N
15.00N
25.00N
35.00N
45.00N
11.81
0
VOUT in Volts
VPOWER in Volts
9.811
7.811
1.000
-1.000
5.811
-2.000
3.811
-3.000
2
5.000N
15.00N
25.00N
35.00N
45.00N
Figure 4, Comparison of the start-up and power supply waveforms with (Top) and
without (Bottom) the new RF bead model.
Page 4
Results
In order to isolate VCC and not contaminate the power supply with
the 500 MHz oscillating waveform, adequate bypassing is required. A voltage generator represents a perfect bypass because
it is zero ohms at all frequencies. This is quite different from the
real world.
As shown in Figure 4, use of an inductor causes a droop in the
VCC voltage. The reason for the droop is that the 170nH represents a large impedance. As the oscillator starts up, the transistor
wants more current. Because of the large inductance it can't draw
adequate current so it starts to discharge the bypass capacitors.
This appears as a drooping in the VCC power (lower graph). The
bead, on the other hand, has a very low DC impedance and a high
AC impedance. By choosing the proper bead, a frequency response can be selected that will block all the AC around the
oscillation frequency. With the bead (upper graph), the VCC line
doesn't droop and shows that the size of the ripple stays the same
revealing the imperfections in the bypassing. Also note that the
oscillation starts slower and does not have quite as much power
out with the bead inserted (to be expected) as it does when the
inductors are used.
The oscillator with no BJT parasitics or beads still oscillates
because it was made to be tolerant of package parasitics, but the
results predicted were inaccurate in several important areas. As
shown in Figure 5, the FFT and transient response of the circuit
with beads and new RF BJT subcircuit model reveals that the
frequency of oscillation is lower and the distortion higher.
Conclusions
From the simulations performed, it is clear that modeling the
proper circuit parasitics is of vital importance, especially at RF
frequencies. It is recommended that initial simulations run for
many cycles in order to verify that stable oscillation is actually
taking place. As for performing an FFT, the ISSPICE .TRAN tstart
parameter can be used to delay the start of data taking until steady
state oscillation has been reached.
With the new bead and BJT models in the RF library, ISSPICE is
able to show the peak component stresses when power is
applied, the transient start-up performance, and the variations in
the power supply. In contrast to the linear analysis programs
commonly used by RF designers, ISSPICE simulations can reveal
many important circuit properties such as efficiency, power dissipation, start up characteristics, and harmonic distortion. Characteristics that would be either difficult or impossible to measure.
Page 5
1.200
800.0M
400.0M
800.0M
1.600
600.0M
With
Beads and
BJT
Parasitics
x 519.5MEG
< 733.3M
>
With
Inductors
and No
Parasitics
400.0M
x 1.052G
< 135.4M
200.0M
>
1
200.0MEG
600.0MEG
1.000G
1.400G
2
1.800G
FFT Responses
2.000
1.000
1.000
3.000
dx = 532.8MEG dy = -597.9M
-1.000
-2.000
2
-1.000
-3.000
91.00N
93.00N
95.00N
97.00N
99.00N
TIME in Secs
Figure 5, Comparison of the frequency spectrum and time waveforms using the new
Intusoft BJT and bead models vs. the Gummel-Poon model and inductor chokes. The
new models give a more accurate prediction of distortion and oscillation frequency.
Page 6
.SUBCKT BEAD 1 2
R4 1 2 220 TC=-.00333
C2 1 2 .9PF
RX 3 2 1E12
CB 3 2 7.432N
F1 1 2 VM1 1
G2 2 3 1 2 1
.MODEL DCLAMP D
E1 4 2 3 2 1
+CJO=8.2286P VJ=25
VM1 4 5
.ENDS
RB 5 2 341.5
RS 5 6 3.7904
VP 7 2 300
D1 6 7 DCLAMP
VN 2 8 300
D2 8 6 DCLAMP
References
[1] Analog & RF Models, Bill Sands (602) 575-5323, FAX (602)
297-5160
[2] RF Device Library, Intusoft, 1990
[3] SIMULATING WITH SPICE, L.G. Meares, C.E. Hymowtiz, Intusoft,
1988
Page 7